2024-09-29-世界银行-通过人工智能驱动的文本挖掘识别扩展的绿色职位清单（英）_33页_847kb

Policy Research Working Paper 10908: Expanded Inventory of Green Job Titles Through AI-Driven Text Mining

Authors: Michał Paliński, Güneş Aşık, Tomasz Gajderowicz, Maciej Jakubowski, Efşan Nas Özen, Dhushyanth Raju
Date: September 2024
Organization: World Bank, Social Protection and Jobs Global Practice

Abstract Summary

This study expands the inventory of green job titles using an AI-driven approach to analyze academic literature published between January 2009 and April 2024, identifying 695 unique green job titles clustered into 25 distinct sectors. The methodology employs a retrieval-augmented generation (RAG) model with natural language processing to enhance reproducibility and scalability in green job categorization, building on O*NET's framework but addressing its limitations of outdated data and U.S.-centric tasks. Results align closely with existing frameworks but introduce new roles, demonstrating AI's potential for refining green economy labor market analyses.

Methodology

• Utilizes a retrieval-augmented generation (RAG) model, specifically GPT-4 with seed parameters, for reproducible job title identification.
• Searches Scopus and Web of Science for articles published after 2008 using keywords related to green jobs and sustainable work.
• Employs text embedding and clustering techniques (e.g., UMAP and HDBSCAN) for job title categorization, focusing on contextual analysis and semantic similarity.
• Conducts robustness checks to validate results, including false negative and positive identifications, with inputs from industry-specific databases and ISIC classifications.

Key Findings

• Identified 695 unique green job titles from 105 articles, with 17% matching O*NET's taxonomy, indicating both alignment and innovation.
• Clusters job titles into 25 sectors, including new roles like green human resources and carbon auditors, reflecting sectors such as renewable energy, construction, and administration.
• Reveals the growing global scope of green job research, with increased geographical diversity from 2009 to 2024.
• Highlights challenges, including publication bias toward high-income countries and limitations in capturing nuanced green task distributions.

Limitations and Implications

• Key constraints include restricted access to full-text articles and binary green/non-green classifications, which may oversimplify job roles.
• Contributes to ongoing green transition discussions by providing an updated, AI-accessible inventory of green jobs.
• Suggests future research should refine AI models, integrate additional data sources, and explore continuous updates to occupational classifications like O*NET.

Conclusion

This study demonstrates the efficacy of AI in expanding green job inventories, supporting evidence-based policies for sustainable labor market transitions.